Fault detection based on belief rule base with online updating attribute weight
Zhijie Zhou, Zhichao Feng, Changhua Hu, Fujun Zhao, Youmin M. Zhang, Guanyu Hu · 2017
In engineering practice, fault detection for complex system is becoming more and more difficult, because enough quantitative observation data can not be obtained. Hence, it is necessary to combine the experts' knowledge and historical data. Belief rule based expert systems have shown excellent performance in modeling complicated relationships with different types of information. However, in current studies, the attribute weights in belief rule base (BRB) are usually determined by experts or system designers. When the engineering environment changes, the attribute weights can not be updated online and this will lose some environment information. In order to solve this problem, this paper aims to propose a BRB model with online updating attribute weight. For the calculation method, the coefficient of variation-based weighting (CVBW) method has been used to calculate the attribute weight and when the new input data are available, the attribute weight can be updated online. A case study for pipeline leak detection has been studied to validate the efficiency of the online updating attribute weight and the experiment has shown that BRB with online updating attribute weight can estimate the leak size and time of pipeline accurately.